Hi, Ive been using HSV and HSL nodes (deactivating channel 1 and 3) to achieve some sort of subtractive saturation feeling to my footage. Working on Alexa 35 RAW files and using ACES CCT on Davinci Resolve 18.6.
I’ve seen this method on a number of YouTube video tutorials and the result is kind of interesting though I’ve found that the image breaks down easily and also that it affects not only the contrast and exposure of it but also, which is more important, the noise or texture artifacts: We’re shooting on a sort of a “film exposure” way, that is to say: exposing for shadows, and the image should be crystal clear. I leave you a couple of images where It’s very noticeable.
Not all types of operations are suited to certain images. And despite optimizing exposure on capture, there isn’t something as a clean camera image. It’s also a hazy background so variance is reduced and you seem to desire saturation from quite low saturated blue tones. This may cause the noise/texture to become visible more quickly than desirable. A solution could be to slightly apply chroma noise reduction. If that doesn’t help perhaps other density style saturation tools/DCTLs may behave better.
Resolve unfortunately doesn’t have smarter/sophisticated tools that garuantee smooth(er) behavior but appying the density to lower exposure and higher saturation ranges more than their opossite is also a solution to naturally exclude the foggy areas. Perhaps that works with a qualifier set to maximum rolloff on the respective channels.
To give you an idea of the kind of modulated operations I’m talking about, Filmlight has developed clever new curve tools for exactly such purposes, diminishing need for qualifiers/keyers.
I have not much experience here but I recently read a final year dissertation describing your issue and providing a solution to fix it. It’s unfortunately in french but perhaps it could still help.
Instead of relying on HSV he uses another color model [1] to apply hue/saturation/value changes which seems behave differently with noise. And he build a DCTL from this (which he doesn’t seems to share the source).
The model is explained at page 86
The tool explanation start page 100
Noise issue showcase page 109
(I do not know if there is alternative model that could work best neither if using various color models is the preferred solution for doing such operation)
Attributes of color represented by a spherical model; CHEN Tieling et al. ↩︎
Gosh, that essay looks dense! I’ll give it a look as it seems to take a deep dive into color concepts… will see if it helps, anyway, the site looks quite a good place to find theoretical resources, thank you.
Ok, after have given a look to the text I see that the author is talking about a tool he has developed to achieve the color saturation trick while avoiding noise plus some sort of green separation tones, both seem quite interesting but I wonder how this could be achieved using the tools that are available, will keep on reading.
Ok, just another update, found that playing around with blur on the HSV node prevents some sort of noise / artifacts, have to be very subtle not to affect the quality of the image, but sort of works.
As you can see in the picture I’ve boosted density wildly while keeping the noise / artifacts controlled… sort of anyways.
Also turned everything into a layer mode and did an alpha node for highlights in the upper node (so to avoid the exposure shift I was noticing)
I would be careful with using spatial operations for a stage of decision making that is about color. While they do go hand in hand in forming the image, blurring a channel in a color model for the sake of better color (per pixel) behavior may not be the best approach. I would keep these separate in the chain. Either denoising the shot itself before going into the look. Or working with spatial tools after the per pixel part of the look.
Have you looked at Mononodes Color Shift? It’s a pretty decent DCTL toolset for manipulating colors in smooth enough fashion. Better than the “tricks” in Resolve I’d say.
Of course, I was just experimenting on Davinci… I read “gaussienne” something in the paper you sent and I had the idea to apply a little blur to the HSL node (S channel), and seemed to work.
Just took a look to mononodes DCTL and reminded me to the color options of photoshop Lightroom to develop RAW files, I use them to try to emulate film stocks… with irregular results I have to admit… its easy to get lost on it.
